Text Prediction Reordering via Random Indexing Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing text input systems for electronic devices rely solely on recency and frequency for ordering predictions, failing to accurately account for contextual relevance, leading to user inefficiency in selecting intended terms.
Innovation Solution
A system utilizing Random Indexing vector space technique to estimate the likelihood of predicted terms belonging within the current textual context, reordering predictions to prioritize relevance, thereby enhancing user input efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If text predictions are ordered solely on the basis of usage frequency statistics, then the system is simple to implement, but the predictions are not accurately relevant to the current textual context
Solution Approach 1:
The patent introduces a vector space model as an intermediary between the text input system and the prediction ranking system. By representing words and contexts as vectors in a multi-dimensional space, the system can compute similarity measures that reflect contextual relevance without requiring complex linguistic analysis. This vector space intermediary enables accurate context-based ranking while maintaining computational efficiency.
Solution Approach 2:
The patent transforms the prediction ranking problem from using simple frequency counts to using vector similarity measures. By changing the parameter from discrete frequency statistics to continuous vector space distances, the system achieves more nuanced contextual matching. The vector representations allow the system to capture semantic relationships and contextual patterns that frequency alone cannot detect.
2Measurement precision
If the system uses immediate lexical context for ordering predictions, then contextual relevance improves, but the system cannot account for nonlocal context
Solution Approach 1:
The vector space model serves multiple functions simultaneously: it captures immediate lexical context through local word vectors, incorporates nonlocal context through document-level representations, and provides a unified framework for both tasks. The same vector space infrastructure handles both short-range and long-range contextual relationships, making the system versatile across different context types without requiring separate mechanisms.
3Ease of operation
If the system displays a limited subset of predictions to the user, then the interface remains simple, but the user may miss relevant terms requiring scrolling
Solution Approach 1:
The patent changes the ranking parameter from simple frequency to vector-based contextual similarity, which fundamentally alters the ordering of predictions. This transformation ensures that the most contextually relevant predictions appear at the top of the limited display, maximizing the utility of the visible subset. The vector similarity measure prioritizes predictions that are semantically appropriate for the current context, reducing the need for users to scroll through irrelevant options.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention relates generally to a system and method for the reordering of text predictions. More particularly, the system and method reorders the text predictions based on modified probability values, wherein th probability values are modified according to the likelihood that a given text prediction will occur within the context of the text inputted by a user.